Complexity-scalable intra-frame prediction technique

[0001] This application claims priority under 35 U.S.C. 119(e) (1) from U.S. Provisional Application No. 60/376,973 filed May 1, 2002.

[0002] The technical field of this invention is that of motion picture data compression and more particularly to coding of intra-frame data.

[0003] This invention concerns coding of motion picture information and relates to the H.26L video standard. Upon reconstruction of the motion picture data some frames, called intra frames, are predicted from adjacent frames. In the current H.26L standard intra-frames are predicted using a nine mode technique. As compared to a prior six mode technique, the nine mode technique is more efficient because it transmits less data. The nine mode technique increases both memory requirement and computational complexity by a factor about four. While the coding efficiency increase is of top priority on the high-end devices, the complexity increase is not favorable for the low-end H.26L applications. Therefore, it would be desirable to have a complexity-scalable intra frame prediction technique that serves the needs of both the low-end and high-end applications.

[0004] In the H.26L standard, during intra frame coding the 16 by 16 pixel luminance part of a macroblock is divided into 16 4 by 4 sub-blocks. Prediction is always used for each sub-block in a macroblock. The prediction mode is determined during encoding. Typically this takes place by comparing the original 4 by 4 sub-block with a corresponding 4 by 4 sub-block formed using each prediction mode. Generally the comparison is made by calculating the sum of the absolute difference of pixels between the original sub-block and candidate predicted sub-block. The prediction mode of the candidate predicted sub-block yielding the lowest sum of absolute differences is selected for that intra frame sub-block. This prediction mode is transmitted to the decoder in the bitstream. Transmission of this prediction mode information may be encoded for data compression as described below.

[0005] FIG. 1 illustrates a 4 by 4 pixel block 100 used to demonstrate prior art technique. Block 100 includes 16 pixels 101 to be coded labeled a through p. Outside and above block 100 are pixels 103 labeled A through H and Q. Outside and to the left of block 100 are pixels 105 labeled I through P. The pixels A to P and Q are from neighboring blocks which are already be decoded and used for prediction. Pixels E through H may not available because they have not yet been decoded, are outside the frame or outside the current independent slice. In this case the value of pixel D is substituted for pixels E through H. Similarly, when pixels M through P are not available, the value of pixel L is substituted for pixels M through P.

[0006] For the luminance signal in the H.26L standard, there are 9 intra frame prediction modes labeled 0 to 8. Mode 0 is DC-prediction mode. The other modes represent directions of predictions as indicated in FIG. 2. The predictions of these modes are given below.

Mode 0:DC Prediction

[0007] All pixels 101 are predicted by the average of the exterior pixels A, B, C, D, I, J, K and L. This is generally calculated as (A+B+C+D+I+J+K+L)/8. If any of the four horizontal pixels A to D are outside the frame, the average of the remaining four is used for prediction. Similarly, if any of the four vertical pixels I to L are outside the frame, the average of the remaining four is used for prediction. If all 8 pixels are outside the frame, the prediction value for all pixels 101 in block 100 is 128. A block will therefore always be predicted in this mode.

Mode 1: Vertical Prediction

[0008] If pixels A, B, C and D are inside the frame, then the pixels a through p are predicted in vertical slices. Thus pixels a, e, i and m are predicted by pixel A; pixels b, f, j and n are predicted by pixel B; pixels c, g, k and o are predicted by pixel C; and pixels d, h, l and p are predicted by pixel D.

Mode 2: Horizontal Prediction

[0009] If pixels I, J, K and L are inside the frame, then the pixels a through p are predicted in horizontal slices. Thus pixels a, b, c and d are predicted by pixel I; pixels e, f, g and h are predicted by pixel J; pixels i, j, k and l are predicted by pixel K; and pixels m, n, o and p are predicted by pixel L.

Mode 3: Diagonal Down/Right Prediction

[0010] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 1.

[0011]

	TABLE 1
	Predicted Pixels	Formula
	m	(J + 2K + L + 2)/4
	i, n	(I + 2J + K + 2)/4
	e, j, o	(Q + 2I + J + 2)/4
	a, f, k, p	(A + 2Q + I + 2)/4
	b, g, l	(Q + 2A + B + 2)/4
	c, h	(A + 2B + C + 2)/4
	d	(B + 2C + D + 2)/4

Mode 4: Diagonal Down/Left Prediction

[0012] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 2.

[0013]

	TABLE 2
	Predicted Pixels	Formula
	a	(A + 2B + C + I + 2J + K + 4)/8
	b, e	(B + 2C + D + J + 2K + L + 4)/8
	c, f, i	(C + 2D + E + K + 2L + M + 4)/8
	d, g, i, m	(D + 2E + F + L + 2M + N + 4)/8
	h, k, n	(E + 2F + G + M + 2N + O + 4)/8
	i, o	(F + 2G + H + N + 2O + P + 4)/8
	p	(G + H + O + P + 2)/4

Mode 5: Vertical-Left Prediction

[0014] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 3.

[0015]

	TABLE 3
	Predicted Pixels	Formula
	a, j	(Q + A + 1)/2
	b, k	(A + B + 1)/2
	c, l	(B + C + 1)/2
	d	(C + D + 1)/2
	e, n	(I + 2Q + A + 2)/4
	f, o	(Q + 2A + B + 2)/4
	g, p	(A + 2B + C + 2)/4
	h	(B + 2C + D + 2)/4
	i	(Q + 2I + J + 2)/4
	m	(I + 2J + K + 2)/4

Mode 6: Vertical-Right Prediction

[0016] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 4.

[0017]

	TABLE 4
	Predicted Pixels	Formula
	a	(2A + 2B + J + 2K + L + 4)/8
	b, i	(B + C + 1)/2
	c, j	(C + D + 1)/2
	d, k	(D + E + 1)/2
	l	(E + F + 1)/2
	e	(A + 2B + C + K + 2L + M + 4)/8
	f, m	(B + 2C + D + 2)/4
	g, n	(C + 2D + E + 2)/4
	h, o	(D + 2E + F + 2)/4
	p	(E + 2F + G + 2)/4

Mode 7: Horizontal-Up Prediction

[0018] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 5.

[0019]

	TABLE 5
	Predicted
	Pixels	Formula
	a	(B + 2C + D + 2I + 2J + 4)/8
	b	(C + 2D + E + I + 2J + K + 4)/8
	c, e	(D + 2E + F + 2J + 2K + 4)/8
	d, f	(E + 2F + G + J + 2K + L + 4)/8
	g, i	(F + 2G + H + 2K + 2L + 4)/8
	h, j	(G + 3H + K + 3L + 4)/8
	l, n	(L + 2M + N + 2)/4
	k, m	(G + H + L + M + 2)/4
	o	(M + N + 1)/2
	p	(M + 2N + O + 2)/4

Mode 8: Horizontal-Down Prediction

[0020] This mode is used only if all pixels A, B, C, D, I, J, K, L and Q are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 6.

[0021]

	TABLE 6
	Predicted Pixels	Formula
	a, g	(Q + I + 1)/2
	b, h	(I + 2Q + A + 2)/4
	c	(Q + 2A + B + 2)/4
	d	(A + 2B + C + 2)/4
	e, k	(I + J + 1)/2
	f, l	(K + 2I + J + 2)/4
	i, o	(J + K + 1)/2
	j, p	(I + 2J + K + 2)/4
	m	(K + L + 1)/2
	n	(J + 2K + L + 2)/4

[0022] Since each of the 4 by 4 pixel luminance blocks are assigned a prediction mode, this requires a large number of bits if coded directly. There are a number of more efficient ways of coding mode information. The best prediction mode of a block is highly correlated with the prediction modes of adjacent blocks.

[0023] FIG. 3 illustrates how the prediction modes of adjacent blocks is used to select the prediction mode of a current block. Referring to FIG. 3, when the prediction modes of block A and block B are known (including the case that either block A or block B or both are outside the frame), an ordering of the most probable, next most probable prediction modes for block C is determined. When an adjacent block A or B is coded by 16 by 16 pixel intra mode, the prediction mode of the current block C is set to mode 0, DC prediction. When an adjacent block A or B is coded by 16 by 16 pixel inter mode, the prediction mode of the current block C is to mode 0, DC prediction in the usual case and outside in the case of a constrained intra update. Table 7 lists this ordering.

[0024]

TABLE 7
B/A	Outside	0	1	2	3
Outside	0--------	01-------	10-------	---------	---------
0	02-------	021648573	125630487	021876543	021358647
1	---------	102654387	162530487	120657483	102536487
2	20-------	280174365	217683504	287106435	281035764
3	---------	201385476	125368470	208137546	325814670
4	---------	201467835	162045873	204178635	420615837
5	---------	015263847	152638407	201584673	531286407
6	---------	016247583	160245738	206147853	160245837
7	---------	270148635	217608543	278105463	270154863
8	---------	280173456	127834560	287104365	283510764
B/A	4	5	6	7	8
Outside	---------	---------	---------	---------	---------
0	206147583	512368047	162054378	204761853	208134657
1	162045378	156320487	165423078	612047583	120685734
2	287640153	215368740	216748530	278016435	287103654
3	421068357	531268470	216584307	240831765	832510476
4	426015783	162458037	641205783	427061853	204851763
5	125063478	513620847	165230487	210856743	210853647
6	640127538	165204378	614027538	264170583	216084573
7	274601853	271650834	274615083	274086153	278406153
8	287461350	251368407	216847350	287410365	283074165

Table 7 lists 9 numbers for each prediction mode of blocks A For example, suppose the prediction mode for block A is mode 2 and for block B is also mode 2. Table 7 lists the string 2 8 7 1 0 6 4 3 5. This indicates that mode 2 is also the most probable prediction mode for the current block C. Mode 8 is the next most probable prediction mode, etc. The compressed bitstream will include information that Prob0=1. This indicates that the next most probable mode (mode 8 in this example) shall be used for block C. The appearance of the symbol ‘B’ in Table 7 indicates that this instance cannot occur because either block A or block B or both are outside the frame.

[0025] For more efficient coding, information on intra prediction of two 4 by 4 pixel luminance blocks are included in one codeword. FIG. 4 illustrates the order of coding of the 16 4 by 4 pixel sub-blocks in a 16 by 16 pixel macroblock. Table 8 lists the universal variable length coding (UVLC) for these combined codewords, where the entries in the UVLC_prob array are the UVLC code_numbers addressed by value of (prob0, prob1) pair UVLC_Prob[prob0=0, 1, Y8] [prob1=0, 1, Y8].

[0026]

TABLE 8
0	1	4	6	10	14	19	25	32
2	3	8	13	17	21	30	37	45
5	9	16	22	27	35	44	46	57
7	12	20	28	36	42	51	59	65
11	18	26	34	38	50	54	63	69
15	23	33	43	49	55	61	68	72
24	29	41	52	56	60	70	74	77
31	39	47	58	62	66	74	76	79
40	48	53	64	67	73	75	78	80

[0027] This prior art intra frame prediction method requires the storage of two tables corresponding to Table 7 and Table 8 in the decoder. The probability Table 7 requires 10×10×9/2=450 bytes memory for an ASIC design. For a digital signal processor (DSP) design this number is higher if each number in Table 7 is stored as byte. This storage requires 900 bytes. The UVLC table (corresponding to Table 8) requires 9×9=81 bytes for storage. Considering that the memory requirement for storing the rest of tables in H.26L standard is less than 338 bytes, this memory requirement for intra frame prediction is extremely high. Such a high memory requirement can be easily critical for ASIC and DSP implementation.

[0028] In low-complexity H.26L codecs, it is desirable to use only a subset of the total nine intra frame prediction modes in order to reduce the memory requirement and computational complexity. However, the current intra frame prediction method cannot be scaled in this manner. In Table 7 prediction modes 0 to 8 spread over all the nine probability positions in each string. Therefore, even if a subset of the prediction modes is used, the decoder has to store a portion of Table 7 and the entire Table 8. For example, if only 6 modes are used, the memory size for DSP implementation is 7×7×9 (portion of Table 7)+81 (UVLC Table 8)=522 bytes. This is barely an improvement over storing all this decode data.

[0029] This invention is a method of encoding intra frames when encoding a motion picture. A set of intra frame prediction modes includes a low-complexity subset. A probability table relates the prediction mode of adjacent sub-blocks to the prediction mode of the current sub-block. For each combination of adjacent sub-blocks, the probability table includes a list of prediction modes in order of expected occurrence. The probability table is adjusted so that each list for prediction modes within the low-complexity subset include initial prediction modes of the low-complexity subset.

[0030] Individual sub-blocks of intra frames are predictively coded as follows. In a low-complexity encoding the prediction mode for each sub-block is within the low-complexity subset. In a high-complexity encoding may use any prediction mode. This permits a low-complexity decoder responsive to only the low-complexity subset.

[0031] The definition of the prediction modes and the probability table and their low-complexity subsets and occurs once upon definition of an encoding standard. Selection of low-complexity encoding or high-complexity encoding may occur once upon manufacture of an encoder or may occur for each encoding of a motion picture.

[0042] The complexity scalable intra frame prediction method of this invention satisfies the needs of both low-complexity and high-complexity H.26L codecs. The memory and computational resource may be highly limited for low-end codecs, so a low-complexity intra frame prediction scheme is desired. On high-end codecs such as PCs, where memory and computational complexity are not a concern, coding efficiency is of high priority. This invention provides a complex and but highly efficient intra frame prediction technique. In addition, there is backward compatibility between the low-complexity and high-complexity techniques. Thus the high-end decoder can decode the bitstreams generated by the low-end encoder. This is achieved because the low-complexity intra frame prediction scheme is a subset of the high-complexity scheme.

[0043] The intra frame prediction technique of this invention operates as follows. Let: {0, 1, . . . M−1) be the M prediction modes supported with probability in descending order; N be the number of complexity scalability layers; M₀, M₁ . . . M_N−2, M_N−1 are number of modes supported in each layer with M₀, <M₁ . . . <M_N−2<M_N−1 (M_N−1=M). Due to the backward compatibility a higher layer contains all the modes supported by a lower layer. Therefore, layer 0 contains mode 0, 1 . . . M₀−1, layer 1 contains mode 0, 1, 2 . . . M₀−1, M₀ . . . M₁−1 . . . and layer N−1 supports all the modes, i.e. mode 0, 1, 2 . . . M−1. The following rules are used for the constructing the probability table shown in Table 9.

[0044] 1. n=0; M₋₁=1;

[0045] 2. For prediction mode of adjacent blocks A and B as illustrated in FIG. 3 less than M, but larger or equal to M_n−1, each string in the probability table is divided into N−n groups: in mode 0, 1 . . . M_n−1 can only be placed from position 0 to M_n−1; mode M_n, M_n+1 . . . M_n+1−1 can only be placed from position M_n to M_n+1−1; . . . and mode M_N−2, M_N−2+1 . . . M_N−1−1 can only be placed from position M_N−2 to M_N−1−1, based on their probabilities; and

[0046] 3. If (n<N−1) {n++, go to 2}; else exit.

[0047] An example illustrated in Table 9 and Table 10 is used to explain the general concept of the complexity-scalable intra prediction described above. In this particular case two complexity layers are established (i.e. M=9, N=2, M₀=6, M₁=9). There are in total 9 prediction modes (0, 1, 2 . . . 8), only six modes (0, 1, 2, 3, 4, 5) are supported in the low-complexity layer. So the nine modes are divided into two sub-groups, layer 0 with modes 0, 1, 2, 3, 4 and 5, and layer 1 with modes 0, 1, 2, 3, 4, 5, 6, 7 and 8. In order to ensure the highest possible coding efficiency in the low-complexity scheme, any mode in layer 0 has higher probability than modes 6, 7, 8.

[0048] A limitation then is imposed on construction of the probability table. As shown in Table 9, for blocks A and B having intra frame prediction modes less than 6 (M₀), the strings are divided into two groups. These are marked in Table 9 by boldface and normal text. Modes 0, 1, . . . 5 can only appear in the first part positions from 0 to 5 (boldface) in order of their probabilities. Modes 6, 7 and 8 can only spread in the second part positions from 6 to 8 (normal text). For blocks A or B having intra frame prediction modes larger than 5, no limitation is imposed. For these cases, the nine modes of the current block can spread all over the nine positions in each string, depending on their probabilities.

[0049]

TABLE 9
B/A	Outside	0	1	2	3
Outside	0--------	021------	102------	201------	012------
0	045------	041253687	104235687	230145678	304215687
1	045------	014253678	102435678	201345678	032154687
2	045------	024135678	120345678	210345678	302145687
3	045------	304152867	310425687	231054687	304215687
4	405------	403512867	401523867	245031687	430512867
5	504------	504312867	015432687	204513687	530412687
6	504------	016234758	261308457	216304578	360214857
7	504------	015463278	102576438	215680347	025136478
8	504------	840315627	038145627	234185607	384156207
B/A	4	5	6	7	8
Outside	012------	012------	012------	012------	012------
0	043152867	045132687	602314587	014627538	048356217
1	041253867	014532678	261354087	160275438	024385167
2	042513687	012543687	261350487	261705348	230451867
3	403512867	304512867	360214587	015364278	380456217
4	405312867	405132867	456231807	401537268	480531267
5	450312867	504132687	560243187	415073286	584631207
6	401362578	145062387	623105487	015462738	034816257
7	064517328	501436278	320156478	102574683	051348726
8	480315627	048351627	304682157	041368257	843025617

[0050] Table 10 lists the universal variable length coding (UVLC) according to this example of the invention for combined codewords according to FIG. 4. Those codes rendered in boldface are required for both low-complexity and high-complexity systems. Those codes rendered in normal type are required only in high-complexity systems.

[0051]

TABLE 10
0	2	5	6	11	15	16	24	46
1	4	9	12	22	25	30	38	59
3	8	14	20	27	36	32	52	63
7	13	21	29	37	43	45	56	70
10	19	28	35	39	47	53	60	71
18	23	34	42	49	55	61	68	76
17	26	33	41	44	54	62	66	75
31	40	50	48	57	65	69	73	79
51	58	64	67	72	74	77	78	80

[0052] Thus the low-complexity intra frame prediction technique uses modes 0 to 5 and only needs to store the sub-tables marked in boldface in Tables 9 and 10. The high-complexity technique stores entire Tables 9 and 10. Since the low-complexity technique is a strict subset of the high-complexity technique, backward compatibility is guaranteed.

[0053] The low-complexity scheme has a reduced memory requirement. In this particular example, the low-complexity decoder requires only 7×7×6 bytes (from the probability sub-table of Table 9)+36 (from the UVLC sub-table of Table 10)=310 bytes to store needed decode data. This is about 40% saving compared to 522 bytes needed in the prior H.26L intra frame prediction when six modes used.

[0054] FIG. 5 illustrates a 4 by 4 pixel block 500 used to demonstrate the invention. Block 500 includes 16 pixels 501 to be coded labeled a through p. Outside and above block 500 are pixels 503 labeled A through D and I. Outside and to the left of block 500 are pixels 505 labeled E through H. The pixels A to H and I are from neighboring blocks which are already be decoded and used for prediction.

[0055] An example nine mode codings corresponding to Tables 9 and 10 above is described below. The (prob0, prob1) coding is unchanged expect for the UVLC table (Table 10) differs from the prior H.26L standard. Note that the modes defined below already existed in the previous H.26L versions. FIG. 6 illustrates the syntax directions of these modes.

Mode 0:DC Prediction

[0056] All pixels 501 are predicted by (A+B+C+D+E+F+G+H)/8. If four of the pixels A through H are outside the frame, the average of the remaining four is used for prediction. If all 8 pixels A through H are outside the frame the predicted value for all pixels 101 is 128. A block will therefore always be predicted in this mode.

Mode 1: Vertical/Diagonal Prediction

[0057] This mode is used only if all pixels 503 A, B, C and D are inside the frame. Pixels a through p are predicted according to Table 11.

[0058]

	TABLE 11
	Predicted Pixels	Formula
	a	(A + B)/2
	e	B
	b, i	(B + C)/2
	f, m	C
	c, j	(C + D)/2
	d, g, h, j, k, l, n, o, p	D

Mode 2: Vertical Prediction

[0059] If pixels A, B, C and D are inside the frame, then: pixels a, e, i and m are predicted by pixel A; pixels b, f, j and n are predicted by pixel B; pixels c, g, k and l are predicted by pixel C; and pixels g, h, l and p are predicted by pixel D.

Mode 3: Diagonal Prediction

[0060] This mode is used only if all pixels A, B, C, D, E, F, G, H and I are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 12.

[0061]

	TABLE 12
	Predicted Pixels	Formula
	m	(H + 2G + F)/4
	i, n	(G + 2F + E)/4
	e, j, o	(F + 2E + I)/4
	a, f, k, p	(E + 2I + A)/4
	b, g, l	(I + 2A + B)/4
	c, h	(A + 2B + C)/4
	d	(B + 2C + D)/4

Mode 4: Horizontal Prediction

[0062] If pixels E, F, G and H are inside the frame, then: pixels a, b, c and d are predicted by pixel E; pixels e, f, g and h are predicted by pixel F; pixels i, j, k and l are predicted by pixel G; and pixels m, n, o and p are predicted by pixel H.

Mode 5: Horizontal/Diagonal Prediction

[0063] This mode is used only if all pixels E, F, G and H are inside the frame. Pixels a through p are predicted according to Table 13.

[0064]

	TABLE 13
	Predicted Pixels	Formula
	a	(E + F)/2
	b	F
	c, e	(F + G)/2
	f, d	G
	i, g	(G + H)/2
	h, j, k, l, m, n, o, p	H

Mode 6: Vertical-Left Prediction

[0065] This mode is used only if all pixels A, B, C, D, E, F, G, H and I are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 14.

[0066]

	TABLE 14
	Predicted Pixels	Formula
	a, j	(I + A + 1)/2
	b, k	(A + B + 1)/2
	c, l	(B + C + 1)/2
	d	(C + D + 1)/2
	e, n	(E + 2I + A + 2)/4
	f, o	(I + 2A + B + 2)/4
	g, p	(A + 2B + C + 2)/4
	h	(B + 2C + D + 2)/4
	i	(I + 2E + F + 2)/4
	m	(E + 2F + G + 2)/4

Mode 7: Vertical-Right Prediction

[0067] This mode is used only if all pixels A, B, C, D, E, F, G, H and I are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 15.

[0068]

	TABLE 15
	Predicted Pixels	Formula
	a	(2(A + B + G) + F + H + 4)/8
	b, i	(B + C + 1)/2
	c, j	(C + D + 1)/2
	d, k	(D + E + 1)/2
	l	(E + F + 1)/2
	e	(2(B + H) + A + C + G + H + 4)/8
	f, m	(B + 2C + D + 2)/4
	g, n	(C + 2D + E + 2)/4
	h, o	(D + 2E + F + 2)/4
	p	(E + 2F + G + 2)/4

Mode 8: Horizontal-Down Prediction

[0069] This mode is used only if all pixels A, B, C, D, E, F, G, H and I are inside the frame. This is a diagonal prediction. Pixels a through p are predicted according to Table 16.

[0070]

	TABLE 16
	Predicted Pixels	Formula
	a, g	(I + E + 1)/2
	b, h	(E + 2I + A + 2)/4
	c	(I + 2A + B + 2)/4
	d	(A + 2B + C + 2)/4
	e, k	(E + F + 1)/2
	f, l	(I + 2E + F + 2)/4
	i, o	(F + G + 1)/2
	j, p	(E + 2F + G + 2)/4
	m	(G + H + 1)/2
	n	(F + 2G + H + 2)/4

[0071] FIG. 7 illustrates a flow chart showing the steps of encoding 700 according to this invention. Encoding 700 begins at start block 701. The initial action of encoding 700 is definition of the intra frame prediction modes (block 702). This definition could be according to the example of modes 0 to 8 described above in relation to FIG. 5 and Tables 11 to 16. This definition of the intra frame prediction modes must include an initial subset of modes suitable for a low-complexity embodiment of the invention. This means that the initial subset must be suitable for a low computational decoding which may yield lower data compression.

[0072] Encoding 700 next defines prediction probability tables for each possible intra frame prediction mode of a predetermined set of adjacent sub-blocks. This could be according to the example of Table 9 above. Each entry in the prediction probability table is a list of the intra frame prediction modes in order of their expected probability. The example of Table 9 is based upon the adjacent sub-blocks defined in FIG. 3. A subset of the probability table in which both the prior sub-blocks are outside the frame or within the subset of intra frame prediction modes of the low-complexity technique must have special properties. Let the number of intra frame prediction modes be N and the total number of modes be M. Each entry in the probability table will have M entries. For probability table entries within the subset, the first N entries of each list must be taken from the subset of prediction modes. As explained above, this permits complete specification of the frame data employing only the subset of intra frame prediction modes. Note that for probability table entries outside this subset, that is probability table entries in which at least one prior sub-block has a prediction mode outside the subset, are not so limited.

[0073] Encoding 700 next defines a universal variable length coding table (block 704). This could be according to the example of Table 10. Note that this universal variable length coding table must include a subset usable with the subset of intra frame prediction modes as described above.

[0074] Encoding 700 next selected the low-complexity technique or the high-complexity technique (decision block 705). The manner of encoding differs depending on this selection as will be described below.

[0075] If encoding 700 operates in the low-complexity technique (LOW at decision block 705), then encoding 700 determines the prediction mode for each sub-block in the current intra frame (block 706). As described above, the prior art technique for this determination includes forming a prediction for each sub-block in each of the prediction modes, then comparing the prediction to the actual sub-block data. The comparison typically is based on the sum of absolute differences between the predicted sub-block and the actual sub-block. The prediction mode yielding the best prediction, such as the lowest sum of absolute differences, is selected. In the low-complexity technique the prediction modes selected are limited to the subset previously defined (block 702).

[0076] Encoding 700 next employs the probability table to encode the prediction mode of the sub-block (block 707). The prediction modes of the adjacent sub-blocks determines the entry in the probability table. The prediction mode determined for the sub-block is matched with the list of the appropriate probability table entry. The determined prediction mode is matched against list of the corresponding entry. The position on this list is noted. In block 708 these positions for two adjacent blocks (as shown in FIG. 4) are encoded together. These paired of probability data is coded via the previously determined universal variable length code (block 709).

[0077] The encoded data for the intra frames is combined with other data concerning the motion picture (block 710). This other data is derived from the original motion picture in a conventional fashion not relevant to this invention. This combined data is transmitted to the decoder via conventional means (block 711).

[0078] If encoding 700 operates in the high-complexity technique (HIGH at decision block 705), then encoding 700 determines the prediction mode for each sub-block in the current intra frame (block 712) in the manner previously described. Note that in the high-complexity technique all prediction modes may be used. This differs from the low-complexity technique which may select only the subset of prediction modes.

[0079] Encoding 700 next employs the probability table to encode the prediction mode of the sub-block (block 713). In contrast to the low-complexity technique, the high-complexity technique may use the entire probability table. In block 714 these positions for two adjacent blocks (as shown in FIG. 4) are encoded together. These paired of probability data is coded via the previously determined universal variable length code (block 715). Note that the high-complexity technique may use the entire universal variable length coding table and not just the subset used by the low-complexity technique.

[0080] The encoded data for the intra frames is combined with other data concerning the motion picture (block 716). This other data is derived from the original motion picture in a conventional fashion not relevant to this invention. This combined data is transmitted to the decoder via conventional means (block 717).

[0081] This invention contemplates that blocks 702, 703 and 704 will be performed once upon setting up the coding standard. Following initialization of the standard, these blocks need not be repeated for every motion picture encoding. This invention contemplates that the low-complexity/high-complexity technique selection of block 705 may be performed upon product manufacture or on initialization of each encoding. An example of selection upon product manufacture is production of an encoder capable of only the low-complexity technique. Such an encoder will always use the low-complexity technique. Another option is production of an encoder capable of only the high-complexity technique. As an alternative, a single encoder could be constructed permitting operator selection of the technique for each motion picture. Such an encoder could encode motion picture data for a low-complexity technique only decoder, for a high-complexity technique only decoder or for a decoder capable of both techniques.

[0082] FIG. 7 fails to provide all the details necessary for a practical implementation of this invention. As an example, FIG. 7 fails to illustrate repetition of blocks 706, 707, 708, 709 and 710 (or blocks 712, 713, 714, 715, 716 and 717) for plural intra frames, nor the encoding of inter frames and other data. However, such details are within the capability of one skilled in the art from this description.

[0083] FIG. 8 illustrates flow chart showing the steps of the low-complexity technique decoding 800 according to this invention. Decoding 800 begins with start block 801. Decoding 800 initially stores the subset prediction modes, probability table and universal variable length code table. An example of the subset prediction modes are modes 0 to 6 described and defined above. An example of subset probability table is illustrated in the boldface entries of Table 9. An example of the subset universal variable length coding table in illustrated in the boldface entries of Table 10. Not that decoder 800 operating using the low-complexity technique can only use the subsets and cannot decode data including modes 6, 7 or 8.

[0084] Decoder 800 next reverses the universal variable length coding of the probability entries (block 803) via the stored table. The two combined sub-block probability indicators are separated (block 804) using the definition of FIG. 4. Decoder 800 next determines the prediction mode for the current sub-block (block 805). The prediction modes of the adjacent sub-blocks determines which list to reference within the subset probability table. The probability indicator for that sub-block determines the place within the list denoting the prediction mode for that sub-block. Decoder 800 employs the thus determined prediction mode and the data from adjacent previously decoded sub-blocks to predicted the current sub-block (block 806). Decoder 800 combines this predicted sub-block data with other data in the received bitstream to form the plural frames of the motion picture (block 807). Decoder 800 enables this motion picture to be displayed or otherwise utilized (block 808). Not shown in FIG. 8 is the necessary repetition for other intra frames.

[0085] FIG. 9 illustrates flow chart showing the steps of the high-complexity technique decoding 900 according to this invention. Decoding 900 begins with start block 901. Decoding 900 initially stores the full data of the prediction modes, probability table and universal variable length code table. An example of the prediction modes are modes 0 to 8 described and defined above. An example of probability table is the whole table listed in Table 9. An example of the universal variable length coding table is the whole table of Table 10. Decoder 900 operating using the high-complexity technique can use any of the modes 0 to 8.

[0086] The remaining portions of decoder 900 are similar to decoder 800. Note that decoder 900 employs all the nine modes, the whole probability table and the whole universal variable length coding table. Thus decoding 900 reverses the universal variable length coding of the probability entries (block 903), separates the two combined sub-block probability indicators (block 904), determines the prediction mode for the current sub-block (block 905), predicts the current sub-block (block 906), combines the other data (block 907) and displays the decoded motion picture (block 908).

[0087] This invention has the following advantages compared to the H.26L intra frame prediction:

[0088] 1. Simplicity: This invention requires only nine boundary pixels from the adjacent blocks (A, B, C, D, E, F, G, H and I in FIG. 5, compare to FIG. 1) in prediction; modes 0 to 5 are much simpler than those in the current H.26L standard.

[0089] 2. Complexity-scalability: This invention allows two layers of complexity scalability. A low-complexity codec can support the first six prediction modes, while a high-complexity codec can use all the nine modes. Full backward compatibility between the high-complexity and low-complexity codecs is maintained. The inventive complexity-scalable intra frame prediction technique provides same coding efficiency as the current nine-mode intra frame prediction technique in a high-complexity codec and as the previous six-mode intra frame prediction technique in a low-complexity codec.